If a video is damaged to a certain extent and polluted by various kinds of noise in a transmission and storage process, the video may lose its essence or be deviated from a requirement of a user. Therefore, video pre-processing becomes an essential means for eliminating bad influence.
A video pre-processing technology refers to a series of operation executed before processing of a video, including image enhancement, image restoration and the like, wherein image enhancement is to add some information or transform data to an original image by virtue of a certain means, and selectively highlight an interested characteristic in the image or suppress some unnecessary characteristics in the image to match the image with a visual response characteristic, thereby improving quality of the image and enhancing a visual effect.
According to different spaces where an enhancement processing process is executed, image enhancement may be divided into two categories, i.e. a frequency-domain-based algorithm and a space-domain-based algorithm. The frequency-domain-based algorithm is to perform certain correction on a transformation coefficient value of an image in a certain transformation domain of the image, and is an indirect enhancement algorithm. For example, the image is considered as a two-dimensional signal, and is subjected to two-dimensional Fourier transform-based signal enhancement. The space-domain-based algorithm is divided into a point operation algorithm and a neighbourhood denoising algorithm. The point operation algorithm refers to gray level correction, gray transformation, histogram correction and the like, and is intended to implement uniform imaging of an image, or extend a dynamic range of the image and extend contrast. The neighbourhood denoising algorithm is divided into image smoothing and image sharpening. Image smoothing is usually configured to eliminate image noise, but may be likely to cause edge blur, and common smoothing algorithms include mean filtering, median filtering and the like. Image sharpening is intended to highlight an edge contour of an object to facilitate target identification, and common sharpening algorithms include a gradient method, an operator method, high-pass filtering, a mask matching method, a statistical difference method and the like.
However, all of the abovementioned image enhancement technologies have the same characteristic that the same enhancement operation is executed on each frame in a video, the adopted image enhancement algorithms are applicable to all the frames in the video, and are undiversified in means and inflexible in processing, enhancement effects may not meet requirements of frames in various scenarios, and particularly when contents of each frame in the video are greatly varied, quality of the video may be deteriorated.